Does mixed frequency variables help to forecast value at risk in the crude oil market?

We construct nine generalized autoregressive condition heteroskedasticity (GARCH) mixed data sampling (MIDAS) models and compare their accuracy in forecasting value at risk (VaR) in the crude oil market with that of benchmark models such as historical simulation (HS) and traditional GARCH models. The main empirical results are as follows. First, we find that mixed frequency information on the demand side of the crude oil market is most helpful for forecasting VaR. Second, although GARCH-MIDAS models generally produce more accurate forecasts than benchmark models, some of the GARCH-MIDAS models tested in this study show poor forecasting accuracy. That is, not all mixed frequency information aids in forecasting VaR in the crude oil market. Third, the HS method, which is widely used by financial institutions, is the least accurate forecasting approach; thus, we do not recommend using HS to forecast VaR in the crude oil market.